Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Qwen3.6-35B-A3B: Agentic coding power, now open to all

qwen.ai

Post reply on HN
Print view View original

321–330 of 563 posts

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#321

Post by gunalx »

lelanthran wrote: post #58
mtct88 wrote: post #9

Nice release from the Qwen team. Small openweight coding models are, imho, the way to go for custom agents tailored to the specific needs of dev shops that are restricted from accessing public models. I'm thinking about banking and healthcare sector development agencies, for example. It's a shame this remains a market largely overlooked by Western players, Mistral being the only one moving in that direction.

> It's a shame this remains a market largely overlooked by Western players, Mistral being the only one moving in that direction. I've said in a recent comment that Mistral is the only one of the current players who appear to be moving towards a sustainable business - all the other AI companies are simply looking for a big payday, not to operate sustainably.

Metawith the llama series as well,they just didn't manage to keep upping the game after and with llama4.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#322

Post by boppo1 »

coder543 wrote: post #165
trvz wrote: post #84

Earlier quoted context omitted.

If you have to ask then your GPU is too small. With 16 GB you'll be only able to run a very compressed variant with noticable quality loss.

Not true. With a MoE, you can offload quite a bit of the model to CPU without losing a ton of performance. 16GB should be fine to run the 4-bit (or larger) model at speeds that are decent. The --n-cpu-moe parameter is the key one on llama-server, if you're not just using -fit on.

I've been way out of the local game for a while now, what's the best way to run models for a fairly technical user? I was using llama.cpp in the command line before and using bash files for prompts.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#323

Post by ilaksh »

tristor wrote: post #85

I'm disappointed they didn't release a 27B dense model. I've been working with Qwen3.5-27B and Qwen3.5-35B-A3B locally, both in their native weights and the versions the community distilled from Opus 4.6 (Qwopus), and I have found I generally get higher quality outputs from the 27B dense model than the 35B-A3B MOE model. My basic conclusion was that MoE approach may be more memory efficient, but it requires a fairly…

It's a given that the dense models with comparable size are better. I also proved that in my use case for those two Qwen 3.5 models.

The benchmarks show 3.6 is a bit better than 3.5. I should retry my task, but I don't have a lot of confidence. But it does sound like they worked on the right thing which is getting closer to the 27B performance.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#326

Post by culi »

simonw wrote: post #270

I've been running this on my laptop with the Unsloth 20.9GB GGUF in LM Studio: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/blob/mai... It drew a better pelican riding a bicycle than Opus 4.7 did! https://simonwillison.net/2026/Apr/16/qwen-beats-opus/

the more I look at these images the more convinced I become that world models are the major missing piece and that these really are ultimately just stochastic sentence machines. Maybe Chomsky was right

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#328

Post by MeteorMarc »

simonw wrote: post #270

I've been running this on my laptop with the Unsloth 20.9GB GGUF in LM Studio: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/blob/mai... It drew a better pelican riding a bicycle than Opus 4.7 did! https://simonwillison.net/2026/Apr/16/qwen-beats-opus/

Interesting, qwen has the pelican driving on the left lane. Coincidence or has it something to do with the workers providing the RL data?

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#329

Post by rdslw »

simonw wrote: post #270

I've been running this on my laptop with the Unsloth 20.9GB GGUF in LM Studio: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/blob/mai... It drew a better pelican riding a bicycle than Opus 4.7 did! https://simonwillison.net/2026/Apr/16/qwen-beats-opus/

interesting, I just tried this very model, unsloth, Q8, so in theory more capable than Simon's Q4, and get those three "pelicans". definitely NOT opus quality. lmstudio, via Simon's llm, but not apple/mlx. Of course the same short prompt.

Simon, any ideas?

https://ibb.co/gFvwzf7M

https://ibb.co/dYHRC3y

https://ibb.co/FLc6kggm (tried here temperature 0.7 instead of pure defaults)

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#330

Post by 0x457 »

zozbot234 wrote: post #193
bigyabai wrote: post #186

Earlier quoted context omitted.

taps the sign Unified Memory Is A Marketing Gimmeck. Industrial-Scale Inference Servers Do Not Use It.

Industrial Scale Inference is moving towards LPDDR memory (alongside HBM), which is essentially what "Unified Memory" is.

> which is essentially what "Unified Memory" is.

Unified memory is when CPU and GPU can reference the same memory address without things being copied (CUDA allows you to write code as if it was unified even if it's not, so that doesn't count, but HMM does count[1])

That is all. What technology is underneath is hardware detail. Unified memory on macs lets you put something into a memory, then do some computation on it with CPU, ANE, ANA, Metal Shaders. All without copying anything.

DGX Spark also has unified memory.

[1]: https://docs.nvidia.com/cuda/cuda-programming-guide/02-basic...

Post reply on HN